blobbers
Let's add some levity to the situation, and here's the link: https://youtu.be/EJR1H5tf5wE?si=lGJ7_L0IN31WNMKE
I ask my agents to go get data from places all the time. Sometimes I ask them to look for undocumented APIs. Is that a bad thing?
Shocked he didn’t end up a crypto maximalist disgraced by FTX ads.
May want to let that marinate.
bzip3 not a terribly impressive or useful compression algo.
benchmark on enwik: https://www.mattmahoney.net/dc/text.html#1703
This should not be surprising at all. Every new students spends tiny fractions of time learning knowledge that took many lifetimes to discover. This fundamental to the progress of intelligence and understanding.
It should not be surprising that AI can be distilled. It's the logical method of training; I would hope that each frontier model is in fact not trained 'from scratch' each time.
We should expect future frontier models are simply distilled versions trained by specialist models, the same way humans learn from a series of professors, papers and canonical books on each different subject material. Models like this can be trained incrementally, or a so called Mixture of Experts (MoE).
When I used to work in cell phones and I was debugging them, when they crash I would analyze the OS core dump for things like repeated phone numbers or emails etc. Basically any thing that signifies a memory leak. But that core dump would be the entire phone’s RAM and storage.
Border guards have always been of questionable value: they inspect your TN-1 and made somewhat arbitrary decisions on your documentation. You’d go one week and they’d send you to secondary but another they’d glance at the docs and just wave you through.
I just felt like the article wasn't particularly well written.
Just because AI can produce slop, doesn't mean humans don't either.
I think slop-pocalypse misses one thing. AI's time of thinking is very different than ours.
A "thick" argument that could take a lifetime in human terms could happen in minutes of AI.
I think to be honest its a good indicator of who I might consider a good boss; if its an uninvolved parent who hires out help to manage their kid and doesn't participate, that's a scary person to work for. They will expect the same of you.
Similarly, those without offspring don't fully understand that free dinner at work isn't a perk. You have a small life form that needs you at home to be present.
Things like kids rolling in to work at 11, eating free lunch, then working a few hours and eating free dinner is no longer fun; it becomes an annoying perception of non-productivity. Partly jealousy, because you used to be able to do that. And now you are regimented around a school and activity schedule.
So while PG may believe that people may have their "act already together" pre-kids, I sort of look at the amount of time people without kids waste and think "if only I had that much free time..."
The back-up model is not a bad idea, but nonetheless...
This seems like the kind of thing that you can have Claude write in an afternoon for whatever service you're running. I don't see the value.
What if we're now moving into a world of strictly KYC. The same way "The Facebook" generated massive revenue by creating a KYC world.
Most of the neural compressors are trying to squeeze every last bit out by sacrificing runtime. You can speed them up a fair bit and they're pretty good. fast-cmix with a small priming corpus will probably be about twice as compact as something like lzma or brotli.
Drug fueled founder parties? Orgies? I've clearly been doing silicon valley wrong. Or right, depending on your perspective... but not this? I've seen board game fueled parties where maybe one or two people do drugs, but it remains very nerdy? And no sex parties. Mostly sausage parties?
... so yeah. Do people really know Jims? Because I just read posts about them.
On a sidenote, is there anything good to replace google news?
Similar thoughts with hackernews...
Chances are if they ghosted you, you didn't do as well as you think. You nailed a tech interview, but you only nailed part 1 of 3 and the interviewer didn't have the heart to tell you there was 2 other questions.
It happens to everyone. We've all done these interviews with nice candidates that just aren't right for that job. Don't worry! You'll find the job that works for you.
I'm curious though if these training methods are convergent or are models actually different; just like how in the stock market people think they're "diversified" but the truth is their exposure is likely much more risk correlated than one might think.
In certain situations, one right answer is better than a committee discussing the problem, but in others its sometimes nice to have some alternative methods of solving something. Fun project nonetheless.
My approach to using multiple models has been less about CoT but more about time to first token, and how you can use a small model to start interacting with the user while in parallel the more complex model is building a larger more complex thought. My work on this was primarily for voice backed interfaces before the voice models became quite a lot faster.
For example, compute X tokens with model A, then feed those into model B, etc. to get chain of thought through a diverse set of mdoels rather than chain of thought through a heterogeneous chain.
Humans seem to strongly believe echo chambers are bad. Are LLMs the same?
Is creatine safe to supplement for children?
Companies like Bloom Energy are putting up power plants right next to DCs. If datacenter power demand then these companies will sell back to the grid creating an oversupply.
If the debt load is too high they're need to declare bankruptcy, at which point the bond holders will lose their capital, not the public.